Microservices architectures conceive an application as a composition of loosely-coupled sub-systems that are developed, deployed, maintained, updated, and scaled independently. Compared to monoliths, microservices speed up evolution and increase flexibility. For these reasons they are becoming the reference architecture for many practitioners. A key challenge to embrace a microservices architecture is how to decompose an application into microservices: a choice that deeply affects all subsequent development phases in ways that are difficult to foresee and evaluate. Without any tool to support their reasoning, developers may erroneously evaluate the various alternatives, leading to inaccurate decomposition choices that would result in increased development, operations, and maintenance costs. This paper tackles the problem with Cromlech, a semi-automatic tool to decompose a software system into microservices. Cromlech (i) takes in input a high-level model of the system in terms of functionalities and data entities accessed by those functionalities, (ii) formulates decomposition as an optimization problem, and (iii) outputs a proposed placement of functionalities and data onto microservices, using a visual representation that helps reasoning on the resulting architecture. Cromlech evaluates design concerns, communication overheads, data management requirements, opportunities and costs of data replication. Our evaluation on a real-world industrial application shows that Cromlech consistently delivers more efficient solutions than simple heuristics and state-of-the-art approaches, and provides useful insights to developers.